A Hybrid Neural Network Approach Based on RNN and CNN for the Detection of Major Depressive Disorder

Depression is a widespread mental illness, impacting over 264 million people globally. It's projected to become the third most prevalent mental illness by 2030. Despite the critical need for accurate diagnosis, traditional diagnostic methods often fall short in terms of precision and accessibility. In contrast, developing a deep learning model to classify Electroencephalo-gram (EEG) signals has the potential to significantly improve depression detection. This research introduces an architecture that integrates the Convolutional Neural Network (CNN) and Gated Recurrent Neural Network Units (GRUs) for classification. CNN layers extract spatial features, which are characteristicsrelated to the distribution and interaction of electrical activity across various brain regions. Subsequently, the GRU examines these patterns to analyze temporal features, which are patterns of signal fluctuation over time, indicative of the brain's rhythmic activity and nonstationary processes. The model was applied toa dataset obtained from the Sains Malaysia Hospital, where 20- channel EEG signals were available from 64 subjects (38 menand 26 women). To verify the performance of the model, five-and ten-fold cross-validations were used. The model demonstrated high effectiveness, achieving an average accuracy of 99.89%, as well as values for precision, recall, and the F1 score of 99.96%, 99.88%, and 99.87%, respectively. These metrics were achieved under defined conditions: a 10-fold cross-validation, a batch size of 64, and 25 epochs. The results highlight the ability of deep learning models to effectively classify depression from EEG data, offering a viable, non-invasive, and economical approach for mental health evaluation.